For most of the web's history, a short link had one job: get a human from a cramped space — a tweet, a poster, an SMS — to a long URL. That job hasn't gone away. But the people clicking are no longer the only ones following your links. AI assistants now read pages, recommend URLs and fetch links on a user's behalf, and they treat links differently from a person with a browser.
This guide covers what has actually changed, and the handful of rules that keep a short link working for both audiences.
Machines follow your links now — and they don't run JavaScript
Every major AI assistant fetches web pages in two ways. A crawler builds an index ahead of time, and a user-triggered fetcher grabs a page the moment someone asks about it. OpenAI documents OAI-SearchBot for ChatGPT search, GPTBot for training and ChatGPT-User for live requests. Anthropic separates ClaudeBot, Claude-SearchBot and Claude-User. Perplexity runs PerplexityBot and Perplexity-User.
What they have in common matters for short links. A large study of crawler traffic by Vercel found that GPTBot and ClaudeBot download JavaScript files but don't execute them. They read the HTML the server sends and stop there.
That makes one class of redirect quietly broken:
- JavaScript redirects (
window.location = ...) do nothing for a fetcher that doesn't run scripts. It sees an empty page. - Meta-refresh and "you are being redirected" interstitials add a page with no content and a delay. At best that's a wasted hop, at worst a dead end.
- Long redirect chains (shortener → tracker → shortener → destination) multiply the chances that something times out or gets blocked.
The fix is boring and reliable: a short link should answer with a single HTTP redirect (a 302/307 response with a Location header) straight to the destination. Every browser, search crawler and AI fetcher understands it.
crisp resolves every short link at the edge with one plain HTTP redirect. No script runs, so a person on a phone and an AI fetcher land on the same destination.
Link previews are a machine audience too
When someone pastes a link into Slack, WhatsApp, iMessage or LinkedIn, a preview bot fetches it to build the card. If your shortener serves those bots its own generic page, the preview shows the shortener's branding instead of your content's title and image.
The better behaviour is to let preview bots follow the redirect to the destination, so the real page's Open Graph tags build the card. The exception is when you've deliberately set a custom preview image for a link. Either way, treat bot visits as previews rather than clicks so your analytics stay honest.
Models make up URLs — so make yours easy to state correctly
Language models generate text one token at a time, and URLs are text. When a model doesn't know a URL precisely, it can produce one that merely looks right. Security firm Netcraft tested this by asking an LLM where to log in to 50 well-known brands. Of the 131 hostnames it suggested, 34% weren't controlled by those brands: 29% were unregistered, parked or inactive, and 5% belonged to unrelated businesses. Unregistered domains are exactly what a phisher would register.
You can't stop a model from guessing, but you can make the correct answer easier to produce and easier to verify:
- Keep canonical URLs short and stable.
crisp.to/launchsurvives being half-remembered, by a person or a model, far better than a 120-character campaign URL. - Publish the canonical link in plain text on pages you control. If an assistant quotes a link to you, you want it copying a real URL from your page, not reconstructing one.
- Use one domain consistently. A brand that scatters links across five shorteners teaches nobody — human or model — which domain is really theirs.
Treat every URL as printed
This is the same discipline as a printed QR code, and the failure mode is identical: the link outlives the thing behind it. See Do QR codes expire? for the print version, and the link durability promise for what crisp commits to.
A link quoted in an AI answer, cached in a chatbot's index or absorbed into a model's training data behaves like a link on a printed flyer. You can't recall it. People will keep asking for it, and models will keep repeating it, long after you've forgotten it existed.
Two consequences follow.
Make links dynamic. The destination should be editable without changing the link. When the campaign page moves, you re-point crisp.to/summer and every copy of it, in emails, in PDFs, in a model's memory, keeps working.
Make sure the domain will outlive the campaign. When Google shut down Firebase Dynamic Links on August 25, 2025, every link on a page.link subdomain started returning HTTP 404, and those subdomains couldn't be kept or transferred. Teams whose links lived on their own domain could point that domain at a new provider. Teams on page.link lost every link at once. If a link might end up somewhere you can't edit, put it on a domain you'll control for years.
301 or 302? Pick the one that lets you change your mind
Redirect codes aren't cosmetic. A 301 Moved Permanently tells clients the move is forever, and browsers may cache it. Change the destination later and some people keep going to the old one. A 302 Found (or 307) is temporary: clients come back to ask each time, so edits take effect immediately.
For dynamic short links, the temporary family is the right default. Search engines and AI fetchers follow both kinds. The difference is only whether you can still change your mind tomorrow.
A checklist for AI-era short links
| Rule | Why it matters now |
|---|---|
| One HTTP redirect, no scripts | AI fetchers and crawlers don't execute JavaScript |
| Temporary redirect (302/307) for editable links | Browsers can cache 301s, which blocks re-pointing |
| Short, readable slugs on one domain | Easier to state correctly; harder to spoof |
| Let preview bots reach the destination | Unfurls show your real title and image |
| Keep bot traffic out of click counts | Analytics reflect people, not crawlers |
| A domain you'll keep for years | Quoted and trained-on links can't be recalled |
| Export and re-point on your schedule | Provider shutdowns shouldn't break your links |
Where this leaves the short link
The short link used to be a convenience. Now it's infrastructure: a stable, machine-readable pointer that you control in a world where your URLs get copied into places you'll never see. The requirements are simple: plain redirects, editable destinations, a durable domain and honest analytics. They just matter more than they used to.